The tuffisite occurs horizontally along the contact between two ignimbrite units, forming a sill of pyroclastic material (Figure 3). The tuffisite fill is composed of a mixture of rhyolitic pumice, dense obsidian and ash shards, together with fragments of the surrounding ignimbrite. The unit beneath the tuffisite and forming its lower boundary is a welded ignimbrite unit composed of massive glass (black ignimbrite), while the unit above the tuffisite and forming its roof is much more friable, fractured, and devitrified (red ignimbrite; Figure 3; Saubin et al., 2019). The black ignimbrite is laterally heterogeneous, appearing less densely welded to the south. Abundant near-horizontal platy fracturing within the red ignimbrite pre-dated tuffisite formation, and facilitated the detachment and incorporation of red ignimbrite blocks into the tuffisite. The wall rock lithology appears to have controlled the morphology of the tuffisite margin, with a sharp planar contact at the tuffisite base, against the black ignimbrite, and a more irregular upper margin as the tuffisite was guided by and exploited weaknesses in the overlying red ignimbrite (Figure 3). Although the upper margin is not planar, the tuffisite does not form offshoots into the roof rock. The tuffisite is offset by two minor faults along its length, displacing it by about 20 cm, and there are numerous sub-vertical fractures through the tuffisite that also cross-cut the overlying red ignimbrite.

The tuffisite has a well-defined but locally complex internal structure consisting of 0.5–20 cm thick units aligned roughly parallel to the tuffisite walls (Figure 4). We adopt a strategy for distinguishing individual depositional units that is based on the following characteristics: grain-size, clast composition, colour, internal structure, and the presence of erosion surfaces.

Most units are laterally continuous for several metres, but thinner units seldom extend >1 m, and such laterally discontinuous units pinch out at the edges after 0.3–1 m. The sequence of units fines upwards from the tuffisite base, but the grain-size increases again immediately above a number of prominent erosive contacts, with material clearly cross-cutting the units underneath. This fining upwards sequence is also interrupted by discontinuous coarse-grained units (Figure 5). All of the units show evidence for a degree of sintering—evidenced by induration—but while some areas show greater sintering, to moderately well-sintered, the degree of sintering does not appear to vary systematically along the tuffisite.

The different characteristics of the tuffisite units allow them to be divided into facies of different grain sizes, described here in order, moving upwards from the tuffisite base in an overall fining upwards sequence. Individual facies are then divided into subfacies of units, defined by the presence or absence of internal structure. Each facies is named according to the pyroclastic lithofacies naming scheme of Branney and Kokelaar (2002) and the characteristics of each facies within the Húsafell tuffisite are summarised in Table 1.

The tuffisite contains a variety of structures that preserve key evidence for emplacement processes. Interpretation of these features will provide constraints on the processes occurring during tuffisite formation, and the fluid pressure required for its emplacement.

The tuffisite consists of multiple units separated by erosion surfaces and, while the units are mostly horizontal, cross-cutting relationships indicate how the tuffisite may have evolved through time. The southwards dip of units in the middle of the tuffisite, along with cross-stratification, suggests that the injected fluid flowed southwards, eroding the underlying material and depositing multiple dipping units (foresets) to produce a structure that migrated laterally over time (Figures 4B, 9). Field evidence suggests that a subsequent pulse of higher velocity fluid eroded the top of this structure, followed by the deposition of the unconformable overlying unit. Shadows behind clasts and soft-sediment deformation in internal injections also suggest fluid flow toward the south. In some areas of the outcrop, the fluid flow direction interpreted from cross-stratification is inconsistent, indicating a range of fluid flow directions. This inconsistency could perhaps be explained by backflow, potentially driven by fluid pressure variations as different fluid flow pathways become blocked, or forward-flow with waxing or waning velocity producing eddies.

Units of varying grain sizes and sedimentary structures inside the tuffisite fill can be interpreted as records of changes in the local velocity of the fluid flowing through the fracture, controlled by spatial and temporal variations in the fluid pressure gradient (Cosgrove, 2001; van der Meer et al., 2009; Phillips et al., 2013). These fluctuations in fluid pressure could be produced externally, controlled by the pressure source, or internally, controlled by processes occurring within the fracture itself.

Volcanic eruptions are rarely steady-state, and so the input of fluid from the unsteady volcanic conduit will be necessarily transient and varying on different timescales. On the longest timescale of days to weeks, the fluid pressure will decrease from a maximum close to the onset of the eruption to lower values as magma discharge wanes. The progressive opening of pathways, for example propagating fractures to shallower depths, will increase fracture volume, lowering the fluid pressure on a timescale of seconds to hours. At the shortest timescale, instabilities within the volcanic conduit, due to unsteady flow, would be able to produce fluctuations in fluid pressure with a timescale of seconds. Fluid pressure fluctuations at each of these timescales will be recorded by the facies and structures of the tuffisite fill.

Even if the source pressure were theoretically constant, internal processes acting within the hydrofracture would still result in fluid pressure fluctuations (Perkins and Kern, 1961). Instabilities in fluid flow can be spontaneously generated within the hydrofracture, producing fluid pressure oscillations that would be superimposed on the fluid pressure variations generated externally. Variations in the width of a hydrofracture along its length could locally alter the fluid pressure (Perkins and Kern, 1961), and the deposition and sintering of particles can clog fluid pathways, lowering the permeability (Farquharson et al., 2017; Heap et al., 2019; Kolzenburg et al., 2019) and therefore changing the hydraulic transport properties of the system. Changes in the fluid pressure gradient along the fracture produce local fluctuations in flow velocity, leading to spatial and temporal changes in (1) erosion when the local fluid velocity is relatively high, (2) deposition when the local fluid velocity is relatively low, and (3) bypassing flow at intermediate flow velocities leading to neither erosion nor deposition. Deposition within the hydrofracture will reduce the space available for fluid flow, locally increasing the fluid pressure until a fluid velocity is reached that can erode the recently deposited sediment (Kern et al., 1959). Once erosion opens more space to accommodate fluid flow the fluid pressure gradient will fall, and so will the fluid velocity, resulting in renewed deposition. In this way, the fluid velocity inside the fracture will always fluctuate. If the source pressure is constant, then these fluctuations may occur around an equilibrium value, at which sediment is not eroded and particles being carried by the fluid are not deposited but transported through the hydrofracture (Kern et al., 1959).

Based on the reasoning explored above, there are three possible origins of the fluid fluctuations recorded by the sedimentary fill within the tuffisite at Húsafell: (1) the source pressure was broadly steady, but instabilities within the tuffisite itself could still produce fluid pressure oscillations; (2) the source pressure was unsteady but waning, leading to fluid pressure oscillations, superimposed on an overall depositional fill; and (3) the source pressure waxed and waned through time, producing a more complex fill formed by deposition then erosion and re-working of sediment that would perhaps contain little systematic variation. The potential magnitude of the pressure fluctuations generated by these different origins is unclear, preventing individual facies changes or structures within the tuffisite from being linked to fluid pressure fluctuations of one particular origin.

Fluctuations in the fluid pressure create the conditions of erosion and deposition needed to produce sedimentary structures, and the presence or absence of structures in different units additionally reflects the particle concentration of the transporting fluid. Some units (facies dslBr, dsLT, and sT) display complex structures such as cross-bedding, laminations, and graded beds, while other units (facies mlBr, mLT, and mT) display less-structured, massive features. These differences can be interpreted as the result of low vs. high particle concentration in the bypassing flow, leading to laminated vs. massive deposits [e.g., indicate deposition from a fluid with a low particle concentration, rather than due to the injection of a slurry with a high particle concentration, which is inferred for massive units (facies mlBr, mLT, and mT; Sparks, 1976; Allen, 1982; Walker, 1984)], or as the result of highly unsteady vs. sustained and steady current conditions at the time of deposition (e.g., Branney and Kokelaar, 2002).

We can combine our observations of sediment characteristics and structures to build an overall model for the emplacement of the tuffisite (Figure 1). The southwards dip of units and sedimentary structures, such as cross-stratification, suggest a roughly southwards direction of fluid flow (Figures 4, 9). The presence of multiple erosion surfaces along the length of the tuffisite is evidence for multiple fluid pulses, which have eroded material beneath and deposited coarser material above (Figure 4B).

Gas and pyroclastic material would have fractured a pathway toward the surface until stalled by the strong and densely welded black ignimbrite, requiring a greater fluid pressure to fracture through. The ascending gas and ash may have reached a great enough fluid pressure to fracture the unit, or exploited an easier pathway, perhaps travelling around the edges of the black ignimbrite unit or fracturing the less densely welded section to the south. Once at the base of the weaker and friable red ignimbrite, the pre-existing sub-horizontal fractures would facilitate horizontal propagation rather than further ascent. As the pre-existing fractures were widened, ignimbrite blocks were detached from the tuffisite walls to produce a single fluid pathway, along which particles were deposited.

The evolutionary model established for the emplacement of the Húsafell tuffisite can be divided into three phases.

To open a hydrofracture, the fluid pressure must be great enough to overcome the lithostatic pressure P_L, induce tensile failure in the surrounding coherent material by overcoming the tensile strength P_T, and widen the hydrofracture by elastically deforming the surrounding material requiring pressure P_W. Inelastic deformation is possible but not considered here as there has been no visible deformation of the fracture walls. The required fluid pressure is therefore

where P_L is calculated using P_L = ρgh, with ρ representing the density of the overlying material, g acceleration due to gravity, and h the depth of the hydrofracture from the Earth’s surface. As injections occur along pre-existing weaknesses such as unit contacts or fractures, the value of P_T is very small compared to P_L and P_W, and so is not considered in pressure estimates below. P_W can be given in terms of the fracture width, W, and fracture length, L, as (Gudmundsson, 1983)

where ν is Poisson’s ratio, and E is the Young’s modulus. If P_Tis negligible, then P_W represents the overpressure (pressure in excess of the ambient P_L) required to open a fracture.

If we assume that the country rock overlying the tuffisite has an average density akin to intercalated lithofacies of pyroclastic, welded, and lava deposits typical of the rhyolitic central volcanoes of Iceland (Ágústsdóttir et al., 2011), we can take an approximate density of ρ = ρ₀(1−ϕ), where ρ₀ is the density of the solid components and ϕ is the average porosity of the overburden. Approximate values might be ρ₀ = 1800kg.m⁻³ for rhyolite, and ϕ = 0.3 for volcanic sequences: ρ = 1260kg.m⁻³. The tuffisite depth below the surface can be estimated as h = 500m via magmatic water concentration in glassy intrusion margins (McGowan, 2016). Taken together, this leads to constraint of P_L≈6.2 MPa.

In order to calculate the P_W component, we must estimate the scales of the tuffisite Wand L, and the properties of the country rock ν and E. Regardless of porosity, the Poisson’s ratio for volcanic rocks is relatively tightly constrained with 90% of available data lying within the bounds 0.10 < ν < 0.35. ν = 0.21 can be found for a tuff with φ = 0.16porosity (Özsan and Akn, 2002), which is also the average of a wide range of measurements using a range of volcanic rocks with porosities from 0.01 to 0.2 (see Heap et al., 2020). Therefore, in this study, we take ν = 0.21. To estimate E, Gudmundsson (1983) assumed that E≈E_d/2, where E_d is the dynamic Young’s modulus, given by

where V_p is the p-wave velocity of the host rock. Estimates of V_p for porous volcanic rock are V_p = 1575m.s⁻¹ (for ϕ = 0.3;Al-Harthi et al., 1999; Vasseur et al., 2016). These constraints lead to E≈1.4GPa. Alternatively, in a review of data for the Young’s modulus of volcanic rocks, Heap et al. (2020) find that the majority of the available data for tuff has an arithmetic mean value of E = 2.4GPa and the majority of the data for tuff materials occur in a moderate-to-high porosity cluster with an arithmetic mean of E = 1.7GPa. This leaves us to find characteristic values of W and L.

The structure of the Húsafell tuffisite is interpreted as a record for fluid pressure fluctuations through time. Using Equation 1 and the constraints provided above, we can estimate the fluid pressure required for tuffisite formation and constrain the maximum overpressure reached by the fluid pressure fluctuations. We compare the pressure required to open the crack hosting the Húsafell tuffisite for two contrasting end-member scenarios. The single-shot model involves the opening of the fracture to maximum width in one pressurisation event, and implies rapid deposition of the whole tuffisite fill width in a single fluid injection event, as a single proppant pulse within a fully dilated (0.9 m-thick and 40 m-long) fracture (Figure 1). The pulsed emplacement model involves sequential injection of a number of thinner units, with multiple sediment pulses within a partially dilated fracture. While the field evidence of continuous margin-parallel deposits implies that the pulsed model still involves a 40-m long fracture, the width of each sequential opening event may be significantly smaller than the tuffisite width.

The single-shot model demands significant elastic deformation of crack walls, and requires an overpressure of 16≲P_w≲29 MPa (lower and upper bound for E = 1.4 and E = 2.4GPa, respectively; Figure 13). Such a high pressure is much greater than the 2 MPa overpressure predicted at 500 m depth for a conduit of width 30 m and an initial water concentration of 4.6 wt% (Degruyter et al., 2012), similar to the ∼5 wt% water measured in Icelandic rhyolites at Torfajökull (Owen et al., 2013). The model by Degruyter et al. (2012) predicts an overpressure of 16–29 MPa only occurs at depths of ∼2 km for a conduit with the same parameters as above. Conduit constriction, not considered by the Degruyter et al. (2012) model, may allow greater gas pressures to be produced at shallower depths; Castro et al. (2016) find that at 300 m depth, reducing conduit width from 400 m to only 25 m can increase the gas pressure by ∼7 MPa. However, even with conduit constriction, a total pressure of 22–35 MPa at 500 m depth is unfeasible. Additionally, such a high fluid pressure would vastly exceed the tensile strength of all country rock lithologies, and be expected to induce significant damage, which is not seen. We therefore conclude that the emplacement of the tuffisite as one single unit requires an unrealistically high gas pressure at 500 m depth, and is therefore untenable as a model.

The pulsed emplacement model requires a more modest P_W compared with the single-shot model, because the pressure required for each incremental injection scales with the partial dilation width (Eq. 2). Many of the tuffisite units, particularly the coarser grained facies (mlBr and dslBr) are visibly continuous across the whole length of the outcrop, and we therefore choose a model unit length of 40 m. Some of the finer-grained tuffisite units are less laterally continuous, but this discontinuity is interpreted as occurring due to erosion, rather than representing the original depositional length of the unit. Field evidence suggests that the tuffisite was deposited in 3 main phases, separated by erosional surfaces (Figure 12). Each phase consists of multiple units of varying characteristics, and often with erosive boundaries. The three phases are therefore interpreted to reflect changes in depositional style as the tuffisite evolves, and are each formed of multiple fluid pulses. The model unit width is taken to be 10 cm, the average width of units of facies mlBR, dslBR, mLT, and dsLT. Units of facies mT and sT are typically thinner than 10 cm and are the least laterally continuous. We suggest that the formation of many of these thinner units is driven by internal fluid pressure fluctuations within the tuffisite, rather than by fluid pressure variations of the source.

Using the range for E given above (E = 1.4 to E = 2.4GPa), injection of a 10 cm-thick, 40 m-long unit, consistent with the emplacement of the entire tuffisite fill in nine successive pulses, would require overpressure of 1.9≲P_W≲3.3 MPa (Figure 13). Attainment of this lower P_W value is far more plausible, and is similar to previous estimates for the overpressure forming tuffisites (Heiken et al., 1988; Saubin et al., 2016) and rhyolitic conduit systems (Benson et al., 2012; Castro et al., 2016). At 500 m depth the model by Degruyter et al. (2012) predicts 2-3 MPa of gas overpressure at 500 m depth for a conduit width of 30 m, consistent with our estimates and our field evidence found here.

We note that the analysis presented above could be repeated, but for a case of a non-porous overlying country rock, for which ρ and V_p, and therefore P_Land P_W, would be higher. However, we note that our estimate of ϕ = 0.3 is typical of a rhyolitic central volcano dominated by pyroclastic sequences.

The single-shot model appears to be infeasible for a thick tuffisite such as that at Húsafell, evidenced by its complex internal structure (Figures 1, 12). The dimensions of the sedimentary units suggest that the tuffisite reached a maximum overpressure of 1.9–3.3 MPa, with fluid pressure fluctuations causing erosion and deposition, producing complex structures. The tuffisite must therefore have formed by pulsatory opening and closing, indicating an unsteady source with fluctuating fluid pressure, though some variation in fluid pressure may be generated by internal processes (see section “A Record of Fluid Pressure Fluctuations”). The overpressure required for tuffisite emplacement, 1.9–3.3 MPa, is similar to the overpressure just above the level of fragmentation, perhaps suggesting that the formation of lateral fractures, able to host tuffisites, is inevitable just above the level of fragmentation. The overpressure estimate is also consistent with the pressure changes inferred from diffusion, with H₂O concentrations in glass-walled tuffisites suggesting transient pressure drops of a few megapascals during fracture opening (Castro et al., 2014).

Field evidence suggests that the finger-shaped injections did not form by fracture opening, subsequent fluid flow and deposition of material, but instead by local distributed deformation of the material above the injection (Figures 4, 11). This is more consistent with the viscous intrusion of a granular medium, fingering into locally deformable surroundings. The finger-shaped injections appear similar to structures produced by the slow intrusion of one granular medium into a second fluid-saturated granular medium, as investigated experimentally in 2D Hele Shaw cells (Saffman and Taylor, 1958; Trevelyan et al., 2011). We draw this analogy between the finger-shaped injections and experimental results in order to suggest that the finger-shaped injections represent a low velocity process that is likely to have occurred toward the end of tuffisite evolution (Phase 3). In turn, this would require that the pressures driving these intrusions were lower than expected via application of the analysis presented in section “Constraints on Tuffisite Emplacement Conditions” (Figure 13).

We note some important differences between the observations of the finger-shaped injections and the processes operative in the 2D Hele Shaw cells (Saffman and Taylor, 1958; Trevelyan et al., 2011). Most importantly, the roof material of the finger-shaped injections is locally partially sintered—a process that cannot be simulated in the low-temperature analogue experiments. This observation also points to a slow, lower-pressure, ductile process rather than the rapid, brittle repeated fracture-opening process invoked for the other units in the tuffisite. Sintering of hot pyroclasts above the finger-shaped injections would be a viscous process, and the pressure required to drive that would depend on the balance between the interstitial gas pressure and the squeezing pressure (Wadsworth et al., 2019). The upper bound on the squeezing pressure is the lithostatic pressure of 6.2 MPa (as discussed in section “Constraints on Tuffisite Emplacement Conditions”), but the gas pressure is unconstrained. Viscous deformation of the roof material that is associated with intrusion injection would allow space for the intrusion to be produced without requiring lithostatic pressure to be exceeded. In a particle-filled fracture the gas pressure is less than lithostatic pressure, but could still be sufficient to deform the overlying material, allowing the finger-shaped injections to form at a relative underpressure, particularly as exceeding lithostatic pressure may cause the overlying material to instead be preferentially lifted to form a fracture. In turn, this is consistent with the finger-shaped injections occurring after the high-pressure fluidised formation of the other units, and thus toward the end of tuffisite evolution, with an overall waning fluid pressure at the source (Phase 3).

The structures inside the Húsafell tuffisite provide a detailed record of the fluid pressure fluctuations during its formation. Elastically opening a space 0.1 m wide and 40 m long for each unit to be injected required 1.9–3.3 MPa of fluid overpressure, and as the fluid pressure waned particles could then be deposited. The fluid overpressure therefore appears to have oscillated, reaching a maximum of 1.9–3.3 MPa during tuffisite formation, with fluid pressure increases allowing for the erosion of previous units. Toward the end of tuffisite evolution the fluid pressure continued to wane, with finger-shaped injections formed at a lower fluid pressure than the previous sedimentary units (<1.9–3.3 MPa).

While some smaller tuffisites do appear to have a simple structure formed by a single fluid pulse (e.g., internal tuffisites at Chaitén; Saubin et al., 2016), the complex structure of the Húsafell tuffisite suggests that it was formed by multiple pulses of material injected into the same fracture. The three phases of deposition described in the model above are the minimum number of injections that occurred during the formation of the Húsafell tuffisite—if the units of each individual phase were each formed by a fluid pulse, there could have been around as many as 20 injections of material into the fracture.

Pulsed injection of pressurised fluid into a hydrofracture (pulsed emplacement model) has been previously suggested in the context of seismic swarms at restless volcanoes (e.g., Chouet, 1996; Kumagai and Chouet, 1999). Swarms of long-period earthquakes (also called low-frequency, and here abbreviated to LP) with very similar waveforms can last for a period of a few hours to several days, with inferred trigger mechanisms involving the repeated excitation of pre-existing cracks. LP events are thought to represent a sudden pressure change within a resonated crack, and may superimpose to create sustained harmonic tremor, which has a common source process that differs only in duration (Chouet, 1996). The quality factor, Q, describes the degree of seismic attenuation, with high Q-values representing long-lasting oscillations. To produce long-lasting oscillations with Q significantly greater than 100, there needs to be a large density difference between the fluid and the surrounding rock (Chouet, 1996). Computed synthetic waveforms for fluid-filled cracks indicate that very high Q-values (e.g., Q = 400 at Tungurahua Volcano, Molina et al., 2004) are best explained if the fluid is a dusty or misty gas with low sound speed (Kumagai and Chouet, 2000; Taguchi et al., 2021). In this volcanic scenario, the dust is inferred to be fine-grained particles of volcanic ash.

At Tungurahua volcano, Molina et al. (2004) modelled resonance of a fracture at 1 km depth, with a length:width (L/W) ratio of 2, length:aperture ratio of 10⁴, and length of ∼200 m, approximately similar to the Húsafell tuffisite. Molina et al. (2004) attributed the systematically changing Q-value during an LP swarm to incremental filling of the fracture by 10 μm ash particles, with eventual crack clogging proposed to have permitted the pressurisation that culminated in an explosive event. This particle size broadly matches that of the Húsafell tuffisite, as well as other documented tuffisite veins for which more detailed particle size analysis has been conducted (e.g., Saubin et al., 2016; Heap et al., 2019). A similar model has been proposed at Galeras volcano, with LP events suggested to represent the propagation and increase in volume of a vertical crack injected by a gas-ash mixture (Taguchi et al., 2018). Both the crack volume and mass fraction of gas within the crack were inferred to have decreased as ash was deposited and welded before the next LP event (Taguchi et al., 2021). We propose that the Húsafell tuffisite represents the fossil record of exactly this type of LP seismic swarm.

Internal tuffisites, which have previously attracted much attention as a potential source of LP earthquakes, are thought to originate in brittle failure events in magma at high strain rates (e.g., Tuffen and Dingwell, 2005; Neuberg et al., 2006). Healing and resealing of these fractures could then permit their repeated reactivation, with a minimum repeat time of a few tens of seconds (Tuffen et al., 2003). However, the modest dimensions of documented internal tuffisites (≤5 m; Tuffen et al., 2003) fall short of the crack dimensions of tens to hundreds of metres in length that are required by crack resonance models (Chouet, 1996; Molina et al., 2004). The small interevent times between LP events are also difficult to explain using the model of magma breaking and healing, unless magma is continually ascending in the seismogenic window (Neuberg et al., 2006; Chouet and Matoza, 2013). We therefore propose that the much more extensive external tuffisite at Húsafell is a better candidate geological record of the LP seismic source process. External tuffisites have the potential to grow to these dimensions, as evidenced by tuffisites at Inyo Domes, California, suggested to extend >120 m from a dyke (Heiken et al., 1988).

Detailed characterisation of seismicity accompanying dyke propagation in Icelandic rift zones, such as at Bárðarbunga in 2014 (Ágústsdóttir et al., 2016; Eibl et al., 2017; Woods et al., 2019), has shown that dyke propagation in the uppermost Icelandic crust (<3 km) is not accompanied by the high-frequency volcano-tectonic (VT) events that are normally characteristic of brittle rock failure. Magmatic pathway propagation is instead thought to involve the widening of pre-existing fractures in weak material (Ágústsdóttir et al., 2016; Woods et al., 2019). Low-amplitude tremor was detected, instead of VT events, and interpreted as a swarm of micro-earthquakes at the propagating dyke tip (Eibl et al., 2017). If we apply this understanding of rift zone seismicity to the structural and lithological context of an Icelandic central volcano then we can speculate on the nature of the seismicity triggered during the opening and lifespan of the Húsafell tuffisite. As the tuffisite initially propagated by widening pre-existing fractures in the weak, overlying red ignimbrite, no high-frequency VT events would have been triggered. Instead, the seismicity would have been low-frequency and solely related to the excitation of pre-existing cracks. However, fracturing of stronger neighbouring formations by the rhyolitic intrusion, such as the underlying conglomerates and black ignimbrite, and overlying basaltic lavas (McGowan, 2016; Saubin et al., 2019), may have involved bursts of seismogenic rupture and small-magnitude higher frequency events. The architecture of diverse country rock lithologies with contrasting mechanical properties therefore guides the spatial distribution and nature of seismicity on the opening and then active magmatic pathway.

The strength of the surrounding tuffisite units must control the location of finger-shaped injections. In a theoretical homogenous tuffisite emplaced in a single pulse, the material closest to the colder country rock should cool most rapidly, and therefore will be less well sintered, weaker, and easier to inject (Kolzenburg et al., 2019). However, in the heterogeneous Húsafell tuffisite, involving emplacement over multiple pulses, the cooling history is more complex. To form a finger-shaped injection the overlying unit requires sufficient cohesion to form a roof, but must maintain the ability to viscously deform at the time of injection, and thus be at high temperature, perhaps aided by additional heat from the injected material. In the Húsafell tuffisite the position of finger-shaped injections will reflect the relative ages and temperatures of material deposited by different fluid pulses. The finger-shaped injections are seen around the centre of the tuffisite width, where slow cooling would allow for the greatest degree of sintering (Kolzenburg et al., 2019), decreasing permeability and allowing fluid pressure to build, and favouring viscous deformation of the tuffisitic roof material. Fluid is injected along the contact between a slow cooled and well-sintered strong unit below, and a younger unit above, deposited by a later fluid pulse, which was still sufficiently hot to viscously deform.

The tuffisite therefore appears to be self-limiting in its lifetime as a degassing pathway. A less-evolved tuffisite with fewer fluid pulses will be colder, allowing material to be more easily injected, and will remain more permeable, allowing for greater amounts of fluid flow. As more pulses are injected and the tuffisite thickens it will take longer to cool, allowing for more sintering to take place, reducing tuffisite permeability. Waning fluid pressure in a hot tuffisite favours the formation of finger-shaped injections, viscously deforming the overlying layers and reducing permeability further.

Tuffisites are a fossil record of the processes occurring during the formation and evolution of magmatic pathways. Quantifying the fluid pressure required to open these fractures, by characterising the mechanical properties of tuffisite host rocks, would inform new models of magma ascent dynamics during pre-eruptive unrest. The field evidence presented here suggests that single-shot models of tuffisite emplacement and associated cooling models are not appropriate, and that a full model for tuffisite emplacement is needed in which fracture width and sedimentation and erosion are coupled in a full dynamic model. Only then could the outgassing flux of tuffisites be properly computed to help assess whether tuffisites can act as pressure release valves capable of modulating the style of silicic eruptions. Finally, swarms of shallow volcanic earthquakes are thought to relate to fluid injection into particle-choked fracture pathways (e.g., Molina et al., 2004). Informing seismic source modelling by using constraints from the tuffisite fossil record could yield improved understanding of the nature of volcanic unrest and its relationship with subsurface magma movement.